DE102019213697A1 - Method for detecting an approach and / or distance of an emergency vehicle relative to a vehicle - Google Patents

Abstract

The invention relates to a method for detecting an approach and / or distance of an emergency vehicle (114) relative to a vehicle (100), the emergency vehicle (114) having a siren (116) for emitting a signal tone (118) and the vehicle (100) a microphone (106) for detecting the signal tone (118). In the method, audio data (112) generated by the microphone (106) are first received in an evaluation unit (102) of the vehicle (100). A cepstrum is then determined from the audio data (112). In the cepstrum, at least one frequency characterizing the signal tone (118) is identified, the course of which over time is recorded. In a further step, a parameter value of at least one statistical parameter is determined for a plurality of successive sections of the temporal course of the frequency in order to obtain a temporal course of the statistical parameter. The approach and / or distance is finally recognized based on the time course of the statistical parameter.

Description

The invention relates to a method, a computer program and a computer-readable medium for recognizing an approach and / or distance of an emergency vehicle relative to a vehicle. The invention also relates to a vehicle system.

A vehicle can be equipped with acoustic sensors to detect ambient noise. By evaluating audio signals from the acoustic sensors, for example, approaching vehicles can be located in the vicinity of the vehicle. The audio signal source can be localized, for example, by means of beamforming. Furthermore, audio signals can be classified by suitable algorithms in order to recognize certain frequency patterns.

The object of the invention is to improve the detection of an audio signal with regard to an approach and / or distance of an emergency vehicle.

This problem is solved by the subject matter of the independent claims. Further embodiments of the invention emerge from the dependent claims and from the following description.

One aspect of the invention relates to a method for detecting an approach and / or distance of an emergency vehicle that has a siren for emitting a signal tone, relative to a vehicle that has a microphone for detecting the signal tone. The method can be carried out automatically by a component of the vehicle, for example an on-board computer, an evaluation unit and / or a vehicle system. The vehicle system can be, for example, an on-board computer or a driver assistance system of the vehicle. It is possible for the vehicle system to be designed to recognize the surroundings of the vehicle and to control the vehicle in accordance with the surroundings, for example to steer, accelerate, brake or navigate. The vehicle system can include an actuator for controlling the vehicle.

An emergency vehicle can, for example, be a police, fire service or rescue vehicle. A siren can be understood as a special signaling system with acoustic and / or optical devices for emitting acoustic and / or optical special signals to warn other road users. A signal tone can be understood to mean an acoustic special signal, for example in the form of a tone sequence made up of a low tone and a high tone or an increasing and decreasing howling tone.

A vehicle can be understood to mean a motor vehicle such as a car, truck or bus or a motorcycle.

A microphone can be understood to be a sound transducer element which is designed to convert sound pressure oscillations into an electrical signal. For example, the microphone can expediently be arranged on a bumper of the vehicle and / or on the side of the vehicle. The vehicle can also have several microphones that can be placed at different points on the vehicle, for example on a front and / or rear bumper and / or on the side of the vehicle and / or on a roof of the vehicle.

According to one embodiment, the method comprises a step of receiving audio data generated by the microphone in an evaluation unit of the vehicle, a step of determining a cepstrum from the audio data, a step of identifying at least one frequency characterizing the signal tone in the cepstrum, a step of recording a temporal progression of the quefrenz, a step of determining a parameter value of at least one statistical parameter for a plurality of successive sections of the temporal progression of the quefrenz in order to obtain a temporal progression of the statistical parameter, and finally a step of recognizing the approach and / or distance based on the temporal course of the statistical parameter.

In other words, the approach and / or distance of the emergency vehicle is recognized by means of an intensity comparison based on a cepstrum without taking into account a direction in which the signal tone is emitted by the emergency vehicle. The audio data generated by the microphone can already have been classified as a signal tone, such as a siren, in advance.

In the context of autonomous driving in particular, it is of interest to differentiate between an approaching emergency vehicle and a retreating emergency vehicle, since an approaching emergency vehicle represents a potential risk to which appropriate measures should be taken. On the other hand, an emergency vehicle moving away represents a lower potential risk. The approach described here makes it possible to make such a distinction with high reliability using vehicle sensors that are already present. In simple terms, a level of received (siren) signals is observed over time and, for example, it is recognized that the corresponding signal source, that is to say the emergency vehicle, is approaching when the level increases, ie when the Siren gets louder. Relevant frequencies can be filtered via the cepstrum, more precisely one or more fundamental frequencies and any overtones of the signal tone. This has the advantage that the influence of disturbance variables such as ambient noise or noise on the detection can be reduced. The development over time of at least one statistical parameter in relation to the cepstrum can also be used advantageously to carry out the detection with the aid of a neural network. This means that new scenarios can be trained flexibly. The detection can accordingly take place reliably.

Audio data can be understood to mean electrical signals generated by the microphone when recording alternating sound pressure vibrations or digital data prepared by suitable further processing of the electrical signals, for example by filtering or transformation.

The evaluation unit can be part of the on-board computer or a control unit of the vehicle system. For example, the evaluation unit can be designed to control the vehicle system. The evaluation unit can be part of the vehicle's on-board computer.

A cepstrum can be understood as a spectrum of a logarithmized frequency range of a signal. A cepstrum is generally generated by an inverse Fourier transform, in which the signal is transformed from the logarithmic frequency range back into a time range. By taking the logarithm in the frequency range, harmonic signal components can be emphasized, even if the signal components only have a relatively small amplitude.

In other words, a cepstrum makes periodic elements visible in the frequency domain of the signal, just as a spectrum makes periodic elements visible in the time domain of the signal.

A frequency can be understood as a variable of a cepstrum function. The Quefrenz is a time value that shows the periods occurring in the spectrum. A frequency characterizing the signal tone can be understood to mean a frequency limit assigned to at least one fundamental frequency of the signal tone and its overtones. The frequency characterizing the signal tone can be determined, for example, by filtering the cepstrum based on one or more filter criteria representing the signal tone.

A statistical parameter can be understood as a parameter for the statistical description and / or analysis of the temporal course of the quefrenz, such as an arithmetic mean, a median, a modal value, a quantile, a standard deviation, a range, an autocorrelation or a combination of at least two of the parameters mentioned.

The successive sections can be time sections each with a constant duration. The sections can also be of the same length, for example. Taken together, the sections can represent, for example, a specific subsection of the cepstrum or the entire cepstrum.

According to one embodiment, the parameter values can be input into a neural network that has been trained for the classes of approach and / or distance. Accordingly, the approach and / or distance can be output as a classifier by the neural network. A neural network can be understood as a network of artificial neurons, whereby the neurons can be arranged in one or more trainable layers lying one behind the other, such as in a convolutional neural network, or CNN for short, which is used for processing image and audio data by machine Learning is suitable. By means of such a classifier, the time profile of the statistical parameter can be assigned to that class to which the time profile of the statistical parameter belongs with the greatest probability. With this embodiment, new scenarios can be learned in a flexible manner. Thus, the recognition accuracy and the robustness of the method can be improved.

According to one embodiment, the approach can be recognized when the time profile of the statistical parameter indicates an increase in the statistical parameter. It is advantageous here that such an increase can be determined with relatively little computing effort.

According to one embodiment, a frequency spectrum can be determined from the audio data by a first Fourier transformation, a logarithmized frequency spectrum can be determined from the frequency spectrum by logarithmizing and the cepstrum can be determined from the logarithmized frequency spectrum by a second Fourier transformation. The first and / or second Fourier transform can, for example, be a fast Fourier transform, or FFT for short. This allows harmonic signal components to be emphasized in the frequency spectrum.

Depending on the embodiment, the statistical parameter can be an arithmetic mean, a standard deviation, a mean slope, an autocorrelation or a combination of at least two of the parameters mentioned.

In this way, interference in relation to the frequency spectrum, such as noise or signal interference, can be compensated.

According to one embodiment, the sections of the temporal course of the frequency limit can each be between 0.1 and 2 seconds, such as between 0.5 and 1.5 seconds, for example 1 second, long. This enables a simple calculation with a sufficiently high resolution.

According to one embodiment, the time course of the frequency and / or the statistical parameter can be smoothed. In this way, random fluctuations in the calculation results can be avoided.

According to one embodiment, the parameter value can be determined for at least ten successive sections of the time profile of the frequency limit. For example, a time series formed by the sections can be 5, 10 or 20 seconds long, depending on the required recognition accuracy. In this way, the time course of the frequency can be clearly determined.

According to one embodiment, the vehicle can be controlled as a function of the detection of the approach and / or distance. Additionally or alternatively, a playback signal can be generated for the acoustic and / or optical playback of a message to a user. A user can be a driver or a passenger, for example. A note can be understood to mean, for example, a warning tone and / or a warning light, for example an illuminated symbol on an instrument panel or the like. For example, when an approach is detected, the vehicle can be steered and / or braked in a suitable manner by controlling one or more actuators of the vehicle, for example in order to make room for the approaching emergency vehicle. In this way, obstructions to the emergency vehicle can be avoided.

Further aspects of the invention relate to a computer program which, when it is executed on a processor, carries out the method as described here, and to a computer-readable medium on which such a computer program is stored. A computer-readable medium can be a hard disk, a USB storage device, RAM, ROM, EPROM or FLASH memory. A computer-readable medium can also be a data communication network, for example the Internet, which enables program code to be downloaded. The computer readable medium can be transitory or non-transitory. Features of the method, as described above and below, can also be features of the computer program and / or the computer-readable medium.

Another aspect of the invention relates to a vehicle system. The vehicle system comprises at least one microphone and an evaluation unit which is designed to carry out the method as described above and below. The vehicle system can in particular implement driver assistance functions. Features of the method, as described above and below, can also be features of the vehicle system and vice versa.

• 1 zeigt schematisch ein Fahrzeug mit einer Auswerteeinheit gemäß einem Ausführungsbeispiel der Erfindung.
• 2 zeigt schematisch einen zeitlichen Verlauf eines Signaltons, eines daraus ermittelten zeitlichen Verlaufs einer Quefrenz und eines geglätteten zeitlichen Verlaufs der Quefrenz zur Auswertung in einer Auswerteeinheit gemäß einem Ausführungsbeispiel der Erfindung.
• 3 zeigt ein Ablaufdiagramm eines Verfahrens gemäß einem Ausführungsbeispiel der Erfindung.

In the following, exemplary embodiments of the invention are described in detail with reference to the accompanying figures.

• 1 shows schematically a vehicle with an evaluation unit according to an embodiment of the invention.
• 2 shows schematically a time profile of a signal tone, a time profile of a quefrenz determined therefrom and a smoothed time profile of the quefrenz for evaluation in an evaluation unit according to an embodiment of the invention.
• 3 shows a flow chart of a method according to an embodiment of the invention.

The reference symbols used in the figures and their meaning are summarized in the list of reference symbols. In principle, identical or similar parts are provided with the same reference symbols.

1 shows schematically a vehicle 100 with an evaluation unit 102 according to an embodiment of the invention. The evaluation unit 102 , shown here as an example of a module of an on-board computer 104 of the vehicle 100 is carried out, is according to this embodiment to two microphones 106 coupled. Depending on the embodiment, the evaluation unit 102 even with more or less than two microphones 106 be coupled. There is a first microphone 106 into a front bumper 108 of the vehicle 100 and a second microphone 106 into a rear bumper 110 of the vehicle 100 built-in. The microphones 106 are designed to record ambient noise and in the form of digital audio data 112 to the evaluation unit 102 to send.

In the in 1 The scenario shown as an example is approaching the vehicle 100 from behind an emergency vehicle 114 with the siren switched on 116 . The siren 116 sends a beep 118 off, for example a tone sequence made up of a lower tone and a higher tone. The microphones pick up accordingly 106 , especially that of the emergency vehicle 114 facing microphone 106 on the rear bumper 110 , the beep 118 , its intensity with decreasing distance from the emergency vehicle 114 to the vehicle 100 increases.

The evaluation unit 102 is executed to from the audio data 112 using suitable transformations to determine a cepstrum in which the harmonic components of the signal tone 118 are reinforced. This makes it possible for fluctuations in the intensity of the signal tone 118 as a result of the approach or removal of the emergency vehicle 114 in relation to the vehicle 100 to be determined without the fluctuations being influenced too much by random disturbance variables. The evaluation unit turns the cepstrum into 102 at least one Quefrenz filtered out that corresponds to the signal tone 118 , more precisely whose fundamental frequency (s) and the associated overtones are assigned. The time course of the Quefrenz is shown in the evaluation unit 102 recorded during a specific recording period and summarized statistically, for example by dividing the recording period into a certain number of equally long time segments and determining values of at least one suitable statistical parameter for each of these time segments. The result of the statistical summary is again a time profile, which this time is formed by the respective parameter values of the statistical parameter used, each assigned to a time segment. The evaluation unit 102 then evaluates the temporal course of the statistical parameter accordingly, as in 1 shown the approach of the emergency vehicle 114 to recognize. For example, the evaluation unit recognizes 102 in this case an increase in the temporal course of the statistical parameter, which corresponds to an increase in the intensity of the signal tone 118 and thus indicates the approach. In the opposite case, the evaluation unit recognizes 102 optionally also a removal of the emergency vehicle 114 , for example when the time course of the statistical parameter indicates a drop in the intensity of the signal tone 118 suggests.

The individual steps of the process, starting with receiving the audio data 112 until the approach or distance is recognized, for example, automatically in the evaluation unit 102 carried out and repeated cyclically.

According to one embodiment, the characteristic values of the individual time segments of the cepstrum serve as input values for a neural network that is fed into the evaluation unit 102 or in another unit of the on-board computer 104 is implemented. For this purpose, the neural network was trained in advance with suitable training data for the “approach” and “distance” classes. On the basis of the characteristic values, the neural network then determines an output value that assigns the input values either to the “Approach” class or the “Distance” class.

According to this embodiment, the microphones 106 and the evaluation unit 102 Components of a vehicle system 120 that is running to the vehicle 100 based on one of the evaluation unit 102 generated evaluation result, here the detection of the approach of the emergency vehicle 114 , by controlling suitable driver assistance functions in a suitable manner, for example to control the vehicle 100 to brake and / or move out of an emergency lane. The vehicle 100 is for this purpose with one or more corresponding actuators 122 like steering or brake actuators. Control of the vehicle 100 is done, for example, by the on-board computer 104 , which sends a control signal in response to the detection of the approach 123 to control the actuator 122 issues.

In addition or as an alternative, there is the on-board computer 104 executed to generate a corresponding playback signal when an approach is detected 124 to create. According to this embodiment, the playback signal is used 124 for reproducing a warning tone via an indoor loudspeaker 126 of the vehicle 100 . The driver of the vehicle is informed by the warning sound 100 in good time before the approaching emergency vehicle 114 warned so that the driver can react accordingly.

2 shows schematically a course over time 200 a signal tone, a time course determined from it 202 a Quefrenz and a smoothed temporal course 204 the Quefrenz for evaluation in an evaluation unit according to an embodiment of the invention, for example in the preceding with reference to FIG 1 evaluation unit described. The history 200 shows, for example, the frequencies of the signal tone contained in the audio data with the unit Hertz over a recording duration with the unit second. The history 202 maps the frequency corresponding to the signal tone over the recording duration. The ordinate represents a dimensionless measure for the intensity, hereinafter also referred to as the cepstral value. The history 202 is wavy, with the intensity steadily increasing within about 15 seconds from the start of recording. The intensity then drops sharply and remains roughly at the level at the beginning of the recording. The history 204 represents the result of smoothing the gradient 202 The wave-like increase in intensity can thus be seen even better. Smoothing is optional.

The smoothed course is used to detect the approach and / or distance of the emergency vehicle 204 according to an embodiment based on a plurality of successive sections 206 of the same duration, as described in more detail below. The sections 206 are in 2 indicated schematically by vertical dividing lines. For example, the entire course 204 or only a certain part of the course 204 in the sections 206 can be divided to perform the detection.

3 shows a flow chart of a method 300 according to an embodiment of the invention. The procedure 300 to detect an approach and / or distance of an emergency vehicle relative to a vehicle, for example, by the evaluation unit 102 how to use 1 is described. The evaluation unit 102 has corresponding modules for this purpose, which can be software components, for example. The software components are installed in the vehicle's on-board computer 100 executed.

Using the microphones 106 an audio signal is recorded first. The audio data becomes the audio data 112 determined, for example by digitizing. The audio data 112 are sent to the evaluation unit and from there in one step 310 receive.

In the evaluation unit takes place in one step 320 the determination of a cepstrum from the audio data 112. For this purpose the audio data 112 first transferred by Fourier transformation into a frequency range, so that the audio data 112 available as a spectrum. Based on the spectrum, the cepstrum is determined. This is done by taking the logarithm of the spectrum and then transforming the logarithmized spectrum back into the time domain.

In the cepstrum is now in one step 330 searched for a cepstral value that represents a relevant fundamental frequency and its overtones. The cepstral value can be understood here as a quantity which indicates a type of amplitude of the frequency (s) sought.

In a subsequent step 340 the cepstral value is recorded over the recording time. This creates a time series 202 as exemplified in 2 is shown, with the time on the abscissa and the cepstral value on the ordinate.

Optionally, the time series can also be smoothed 202 to eliminate random fluctuations. The course over time 204 in 2 shows the result of such a smoothing as an example.

In order to be able to predict at a certain point in time, approximately after 10 seconds, whether the emergency vehicle will be 114 approaches or removes, becomes a history of time periods 206 , for example with a length of 1 second each, based on a selected starting time.

In these (relatively short) periods of time 206 be in one step 350 various statistical properties such as mean value, standard deviation, mean slope or autocorrelation, also called statistical parameters, determined.

Based on a temporal development of the statistical properties, in one step 360 for example, determines that the mean value is growing. The approach of the emergency vehicle is corresponding 114 , ie the source of the beep 118 , recognized.

According to one exemplary embodiment, the development over time of the statistical parameters over the time segments 206 selected as the input variable for a neural network that was trained for the classes “approach” and / or “distance”. Accordingly, the approach or distance from the neural network is output as a classifier, by means of which the development of the statistical parameters over time is recognized either as an approach or as a distance.

It is also possible, as described above, that the vehicle 100 in an optional step 370 depending on one in the crotch 360 generated recognition result is controlled accordingly. Additionally or alternatively, this is done by generating a corresponding playback signal 124 a user of the vehicle 100 pointed to the recognition result.

In addition, it should be pointed out that “comprehensive” does not exclude any other elements or steps and “one” or “one” does not exclude a large number. Furthermore, it should be pointed out that features or steps that have been described with reference to one of the above exemplary embodiments can also be used in combination with other features or steps of other exemplary embodiments described above. Reference signs in the claims are not to be regarded as a restriction.

List of reference symbols

100

vehicle

102

Evaluation unit

104

Board computer

106

microphone

108

front bumper

110

rear bumper

112

Audio data

114

Emergency vehicle

116

Sirloin

118

Beep

120

Vehicle system

122

Actuator

123

Control signal

124

Playback signal

126

Indoor speakers

200

temporal course of the signal tone

202

temporal course of the quefrenz

204

smoothed temporal course of the Quefrenz

206

Sections of the course of time

300

Method for detecting an approach and / or distance of an emergency vehicle relative to a vehicle

310

Step of receiving

320

Step of determining

330

Step of identification

340

Step of recording

350

Step of determining

360

Step of recognition

370

Step of controlling and / or generating

Claims (12)

Method (300) for detecting an approach and / or distance of an emergency vehicle (114) relative to a vehicle (100), the emergency vehicle (114) having a siren (116) for emitting a signal tone (118) and the vehicle (100) on A microphone (106) for detecting the signal tone (118), the method (300) comprising: Receiving (310) audio data (112) generated by the microphone (106) in an evaluation unit (102) of the vehicle (100); Determining (320) a cepstrum from the audio data (112); Identifying (330) at least one frequency characterizing the signal tone (118) in the cepstrum; Recording (340) a history (202, 204) of the frequency limit; Determining (350) a parameter value of at least one statistical parameter for a plurality of successive sections (206) of the temporal course (202, 204) of the frequency in order to obtain a temporal course of the statistical parameter; Detecting (360) the approach and / or distance based on the time course of the statistical parameter. Method (300) according to Claim 1 , wherein the characteristic values are input into a neural network that has been trained for the classes approach and / or distance, and the approach and / or distance are output as classifiers by the neural network. Method (300) according to one of the preceding claims, the approach being recognized when the time course of the statistical parameter indicates an increase in the statistical parameter. Method (300) according to one of the preceding claims, a frequency spectrum is determined from the audio data (112) by a first Fourier transformation, a logarithmized frequency spectrum is determined from the frequency spectrum by logarithmizing and the cepstrum is determined from the logarithmized frequency spectrum by a second Fourier transformation. Method (300) according to one of the preceding claims, wherein the statistical parameter comprises an arithmetic mean, a standard deviation, a mean slope and / or an autocorrelation. Method (300) according to one of the preceding claims, wherein the sections (206) of the temporal course (202, 204) of the Quefrenz are each between 0.1 and 2 seconds long. Method (300) according to one of the preceding claims, wherein the time course (202, 204) of the frequency and / or the statistical parameter is smoothed. Method (300) according to one of the preceding claims, wherein the parameter value is determined for at least ten successive sections (206) of the temporal course (202, 204) of the frequency limit. Method (300) according to one of the preceding claims, wherein the vehicle (100) is controlled as a function of the detection of the approach and / or distance and / or a playback signal (124) for acoustic and / or optical playback of a message to a user is generated. Computer program which, when executed on a processor, carries out the method (300) according to one of the preceding claims. Computer-readable medium on which a computer program is based Claim 10 is stored. A vehicle system (120) comprising: a microphone (106) for detecting a beep (118); an evaluation unit (102) which is designed to carry out the method (300) according to one of the Claims 1 to 9 perform.

Images